Recent research conducted by Icertis has revealed significant visibility challenges facing in-house legal teams regarding the actions of AI agents. Close to half of surveyed U.S. corporate legal professionals admitted they would only detect unauthorized or incorrect AI actions well after they occurred, sometimes days or weeks later. This stems from a survey encompassing more than 1,000 practitioners, highlighting a governance gap accompanying the increased autonomy of AI tools.
Despite the predominant use of AI in supportive roles, approximately 25% of respondents allow AI to autonomously handle certain tasks. Notably, nearly 10% of practitioners indicated that human oversight over AI activity is minimal. Nevertheless, only 23% of legal teams have a comprehensive AI policy specifically for agentic AI, with 60% expressing belief that their governance frameworks will be apt for overseeing AI agents within the next 12 to 24 months.
Concerns regarding the accuracy of AI outputs add to the complexity of governance, with only 26% expressing high confidence in the AI’s reliability for critical business decisions, and about 50% relying on human judgment to validate AI outputs. Furthermore, the survey identified fragmented visibility, as only 39% feel confident about having real-time oversight over AI agents, matched by those who believe they would only identify issues post-incident.
Accountability for AI mistakes appears to be a contentious issue. Opinions were divided, with nearly equal percentages indicating responsibility lying with deployment teams, operational managers, or context-dependent scenarios. Interestingly, although legal departments are often primary custodians of AI policies, only 10% felt they would shoulder responsibility in compliance breaches.
Regarding data integration, the survey indicated that over 70% use generic language models like ChatGPT or Claude, particularly in contract-related tasks. However, only a fraction have AI tools capable of bi-directional data exchange with other business systems.
Icertis proposes contracts as a governance layer as a potential solution, suggesting that contract data can enrich AI with necessary business context to improve decision accuracy. Despite this, only 38% currently consider contracts for AI governance, though there is some recognition of their potential utility.
According to Bernadette Bulacan, chief evangelist at Icertis, the rapid evolution of AI is outpacing its regulatory oversight, exerting pressure on legal teams both internally and through increased AI use in other organizational areas.
The complete findings and analysis can be accessed via the Icertis website.